adjusted vicuna template

sotopia-lab · Nov 1, 2023 · d70dce7 · d70dce7
1 parent b76fc0f
commit d70dce7
Show file tree

Hide file tree

Showing 2 changed files with 5 additions and 5 deletions.
diff --git a/llm_ft/fastchat/conversation.py b/llm_ft/fastchat/conversation.py
@@ -983,10 +983,10 @@ def get_conv_template(name: str) -> Conversation:
 register_conv_template(
     Conversation(
         name="sotopia-llama-2",
-        roles=("Agent1", "Agent2"),
-        sep_style=SeparatorStyle.LLAMA2,
+        roles=("PROMPT", "RESPONSE"),
+        sep_style=SeparatorStyle.ADD_COLON_TWO,
         sep=" ",
-        sep2=" </s><s>",
+        sep2="</s>",
     )
 )
 

diff --git a/llm_ft/llama2-13b_qlora_train.sh b/llm_ft/llama2-13b_qlora_train.sh
@@ -3,7 +3,7 @@ deepspeed --num_gpus=1 fastchat/train/train_lora.py \
     --lora_r 8 \
     --lora_alpha 16 \
     --lora_dropout 0.05 \
-    --data_path ./data/fastchat-ft-gpt4-gpt4-easy-2-side-partial.json \
+    --data_path ./data/fastchat-ft-gp4-gpt4-easy-truncated.json \
     --shuffle True \
     --bf16 True \
     --output_dir ./checkpoint-shuffle \
@@ -25,7 +25,7 @@ deepspeed --num_gpus=1 fastchat/train/train_lora.py \
     --hf_access_token "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" \
     --tf32 True \
     --flash_attn True \
-    --template "llama-2"
+    --template "sotopia-llama-2"
 
 # Possible other options
 # --flash_attn True \